LLM-Powered Browser Automation Agent
Published:
A production-ready browser automation agent pairing an LLM reasoning loop with Playwright to autonomously complete multi-step web forms.
Published:
A production-ready browser automation agent pairing an LLM reasoning loop with Playwright to autonomously complete multi-step web forms.
Published:
Fine-tuned ParsBERT for binary sentiment classification on Persian-language customer reviews.
Published:
A hybrid recommendation pipeline combining collaborative filtering, content-based features, and Pearson correlation.
Published:
An end-to-end RAG framework combining DiCE-generated counterfactual candidates with LLM-based refinement to improve the feasibility and plausibility of recourse recommendations.
Published:
An uncertainty-aware Bayesian model that infers a person’s values from conversation, behavior, and reflection, with counterfactual previews for reflection before a decision is made. Ongoing research, Augmented Intelligence Lab, University of Waterloo.
Published:
LSTM and GRU models forecasting 30-day stock closing prices, comparing performance across architectures.
Published:
An end-to-end detection pipeline achieving 95.4% image-level accuracy and F1 = 0.967, with real-time drift monitoring.
Published:
A deep learning ensemble of neural networks for road sign recognition.
Published in B.Sc. Thesis, Shahid Beheshti University (under review as a preprint), 2023
A protein toxicity prediction module combining transformer-based protein embeddings with a tuned LightGBM classifier, awarded Best B.Sc. Thesis.
Recommended citation: Zahra Khotanlou. "ToxTrans: A Deep Learning Method to Predict Protein Toxicity." B.Sc. Thesis, Shahid Beheshti University. Supervisor: Dr. Hamed Malek.
Published in Preprint, under review, 2026
A reproducible benchmarking platform integrating 22 heterogeneous counterfactual-explanation methods across 137 experimental configurations.
Recommended citation: Zahra Khotanlou, Hashir Ahmed, Chenghao Tan, Ahmed Abdelaal, and Amir-Hossein Karimi. "RecourseBench: Reproducible Counterfactual Explanation Evaluation." Preprint, under review.
Download Paper
Published in ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026), 2026
A many-to-many algorithmic recourse framework modeling interactions among users and decision providers under shared capacity constraints, formulated as a capacitated bipartite matching problem.
Recommended citation: Zahra Khotanlou, Kate Larson, and Amir-Hossein Karimi. (2026). "Your Recourse, My Loss? Algorithmic Recourse under Shared Constraints." ACM FAccT 2026.
Download Paper
Published:
Presented at an interdisciplinary AI and Neuroscience session hosted by the Sharif Cognitive Science Community.
Published:
Presented findings from my ACM FAccT 2026 paper on multi-agent algorithmic recourse and capacity-constrained recourse allocation.
Undergraduate courses, Shahid Beheshti University, 2021
Teaching Assistant across seven undergraduate CS courses: Machine Learning, Artificial Intelligence, Advanced Programming, Fundamentals of Computer and Programming, System Analysis and Design, Linear Algebra, and Embedded and Real-time Systems.
Graduate courses, University of Waterloo, 2025